How to extract specific edges from an image?

Hello eveyone, I have an image of a car door from which I have to find generate a trajectory from specific edges of the car door. Let me expain the detail about my project.

Project: Finding the edges of a door and used those edges as a trajectory for a sensor that checks edges for Gap/Flush.

Principle: I used Savitzky Golay, Edge Linking and Line Segment Fitting (https://www.peterkovesi.com/matlabfns/#step2line).

Method I used: Image Analyst's Savitzky filter smoothening at three diffrent thresholds, obtain three images and blend them. The output is again processed to Edge linking and line segmenting program.

After all this I am not able to get good edges, its either not smooth or has a lot of unnecessary edges. But the main problem is I am getting a lot of unnecessary edges in the image. To get a Trajectory I need only Three specific edges which I have marked by hand below in red.

Questions:

1. How to remove unnecessary edges out of the image. Please suggest me a way to get only those three edges? (without manual intervention)
2. How to smoothening those edges and join them so that i can use it as trajectory for another program?

I have attached the images of input and output. Sorry for the long question.

Answer 1

[Image Analyst]

If you have an edge image with a bunch of edges but want edges from only certain locations, you maybe best off using a mask/template to erase everything outside the areas you want to look in.

[Harshan Desu]

Hello Sir, thanks for your support. Your answers and programs have been very useful for me as a beginner.

I used a template/mask to erase everything else and obtained the edges I needed.

Now how do I extract edges and curve fit to get a function so that I get the Trajectory of the above curve. This is my code..

    im = imread('blend2_21_OPPFS.png');
    im = rgb2gray(im);
    props = regionprops(im, 'Area');
    %im = imread('pout.tif');        % data

% smooth the image
clc;
clear;
close all;
% Read Input Image

InputImage=imread('blend2_22_OPPFS.png');

%Resize the Image

%InputImage=imresize(InputImage,[256 256]);

% Display the Image

imshow(InputImage);

% Get Inputs from Mouse,Select 4 Seed Points in Image

[Col Row]=ginput(6);
c =Col;
r =Row;

%c = [352.313364055300;100.645161290323;119.571428571429;132.870967741936;353.847926267281;353.847926267281];
%r = [111.244239631336;82.5990783410139;140.912442396313;181.322580645161;188.483870967742;187.460829493088];

% Select polygonal region of interest
BinaryMask = roipoly(InputImage,c,r);
figure, imshow(BinaryMask);title('Selected Region of Interest');

%Create Buffer for ROI
ROI=zeros(333,476);

%Create Buffer for NONROI
NONROI=zeros(33,467);
for i=1:333

for j=1:476

if BinaryMask(i,j)==1
ROI(i,j)=InputImage(i,j);

else
NONROI(i,j)=InputImage(i,j);
end

end

end

%Display ROI and Non ROI
figure;
subplot(1,2,1);imshow(ROI,[]);title('ROI');
subplot(1,2,2);imshow(NONROI,[]);title('NON ROI');

[Harshan Desu]

Had to replace door with another component called Cup holder of car. I have worked and got all thw edges sir. Now all I need is two edges, How do I filter all the other edges and get only the two edges I wanted?

[Image Analyst]

There are lots of edges in that image. How do you want to indicate the two that you want?

[Harshan Desu]

Hi, sorry for being vague Sir, had to do a free hand sketch.

The First edge is as I marked, the all round edge

The second edge is as I marked, it can also be all around like the first edge.

Once again Thank you for taking time!

With Gratitude,

Harshan D

[Image Analyst]

I don't know. I'd probably suggest you try segnet deep learning on the original grayscale or RGB image.

[Harshan Desu]

Oh okay Thanks anyway. I started searching in Matlab Forum and google sir. It seems very complex and scary sir. Can u suggest any examples or references that I can use as a Novice.

[Image Analyst]

No, this might be too tough a problem for a novice. If you can, you might want to start learning with easier challenges.

[Harshan Desu]

Is there any other way sir, I just want the 2 curves and rest all dissapeared.

For the earlier door image I thresholded 3 times with different values to get 3 specific curves. But same Threshold alongside Savitzky Golay filter is not doable for new image as the Binary image is just full black after threshold = 220.

[Image Analyst]

Try a threshold that gets all the edges then use bwselect() to manually pick out the one you want.

[Harshan Desu]

The enitre process has to have no manual interverntion. Thanks for the Idea anyways!